Xero · How it works
How Cosmic Bills reads a supplier invoice
Optical character recognition, a trained extraction model, and a review step. What actually happens between uploading a PDF and a coded bill in Xero.
Published · Updated
Xero · How it works
Optical character recognition, a trained extraction model, and a review step. What actually happens between uploading a PDF and a coded bill in Xero.
Published · Updated
"It reads your invoices" is a description that does not tell you much about what to expect. Here is what happens between dropping a PDF onto the page and seeing a coded bill in Xero.
Optical character recognition converts the PDF into text with positions attached — not just what the words are, but where on the page they sit. Layout matters, because on an invoice the meaning of a number is largely determined by which column and row it is in.
A trained model works out which of those values is the supplier, the invoice number, the dates, the tax, the total, and — the difficult part — where the line-item table starts and ends and what each of its columns means. Supplier invoice layouts vary enormously, which is why this is a learned task rather than a set of rules.
The supplier is matched against your contact list, and the account coding you have configured for that supplier and its products is applied. This step is why the second month is better than the first.
The result appears next to the original document. You can edit any value, add or delete lines, split a line across cost centres, and then approve. Nothing posts to Xero until you do.
Step four is not optional
Of everything a finance team does, accounts payable is high-volume, rule-driven and slow to optimise. That is why it goes first.
Scan history, PDF preview, supplier mapping and line splitting — the parts people actually use every week.
Manual entry of a dozen multi-line bills is slow enough to be worth measuring. Here is how to work out what yours costs.
Start on the free plan, connect Xero or Reckon, and process your first bills today.